Why Some AI Leaders Want to Slow Down New AI
Anthropic CEO Dario Amodei has proposed to "pace the frontier" of AI development, in a blog post built around three broad ideas. TechCrunch
The plan centers on outside checks and teamwork. Independent safety testers would check the newest and most powerful AI systems, and AI labs in democratic countries would work together. TechCrunch
OpenAI CEO Sam Altman agreed with Amodei that it is time to "pace the frontier." TechCrunch The idea has some wider support in the industry, and pushback from Jensen Huang. TechCrunch Huang is the chief executive of Nvidia, which makes AI chips. The New York Times
That public debate runs alongside quieter talks between labs. OpenAI, Anthropic and Google have been in talks on AI safety for weeks. TechCrunch Google DeepMind has launched an institute to widen the debate over AGI, the idea of very capable future AI. TechCrunch
On TechCrunch Equity, hosts Kirsten Korosec, Anthony Ha and Sean O'Kane took up the practical question behind the slogan. They discussed whether companies can agree on what slowing down means and who gets to police it. TechCrunch
The broader context here is simple to state and hard to solve. Saying the goal is easy. Agreeing on it is harder. Outside testing only works if testers get full access, clear tests and enough time. Like building inspectors, they need to go inside, not just look from the street. That means the inner workings of the model, helper software, tool use and how it will be used, not just limited access for a few days. It means repeat tests that still work when wording or later training change slightly. And it means testing before a release is locked in, while a bad result can still delay launch.
In my view, teamwork is the harder problem. Labs in democratic countries still compete for workers, computers, business deals and users. A voluntary deal on pace would ask rivals to share safety findings, accept outside judgment and delay releases that could win customers. Worth flagging as an open issue is checking. Without a way to confirm what was built, tested and shipped, teamwork rests on trust between rivals.
In practical terms for people using these models, the near-term effects are about process, not new technology. If outside review becomes normal, leading teams will need test-ready versions, plain records and emergency plans as routine work. If lab talks hold, the safety talks among OpenAI, Anthropic and Google could become a regular meeting rather than a one-time response. If not, each lab will keep its own rules and ask the public to trust its own tests.
My own expectation, shaped by watching my children grow up through desktop to mobile to cloud, is that use will not wait for rules to feel settled. Developers will build on whatever new models exist. That makes clear tests, shared results where possible and common language for risk useful. The proposal does not settle the debate. It gives it a structure to work with.
